BACKGROUND Precise evaluation of the haemocompatibility of prototype membranes, flow configurations and anticoagulant regimens is an essential step in the development of dialysis systems minimizing blood activation. An ex vivo model in humans currently employed in our laboratory has recently been adapted to allow the parallel evaluation of two minimodule dialysers with blood from a single donor, thus eliminating differences due to donor variability in the comparison of test and control dialysis modules. METHODS The ex vivo flow system is designed to reproduce the haemodynamic conditions of clinical dialysis on a 1/50 scale. A blood line from the forearm vein of the volunteer donor is divided at a Y-shaped junction, two roller pumps assure equivalent blood flow (5 ml/min) in the branches leading to two minimodule dialysers and heparin (0.1 IU/ml final concentration) is injected into each branch immediately after the Y junction. Samples for analysis of blood activation markers are collected at the exits of the two minimodules over a test period of 27 min. In the present series of tests, a new polyacrylonitrile membrane (PAN) was evaluated relative to standard commercial polysulphone (PS), acrylonitrile copolymer (AN 69) and cuprophan (CUP) membranes. RESULTS A steady minimal level of anticoagulation corresponding to a slightly less than two-fold prolongation of APTT (activated partial thromboplastin time) was maintained throughout testing in both branches of the ex vivo flow system. Time curves for the accumulation of activation markers (thrombin-antithrombin III complexes, prothrombin fragment 1 + 2, platelet beta-thromboglobulin, and complement fragment C3a) showed all four types of minimodule dialyser to induce comparable low levels of activation of coagulation parameters and platelets, together with similar mild activation of complement for AN 69, PAN, and PS dialysers as compared to stronger activation for CUP modules. Overall results thus confirmed the acceptable haemocompatibility of the prototype polyacrylonitrile (PAN) membrane. CONCLUSIONS Among current methods for evaluation of the biocompatibility of haemodialysis systems, ex vivo flow models in humans avoid problems arising from species differences and may be designed to closely reproduce the conditions of clinical dialysis. A parallel configuration eliminates artefacts due to individual variations in donor response. This not only facilitates the direct comparison of test and control membranes under close to identical experimental conditions, but also provides a model particularly well adapted to studies of the effects of different anticoagulation regimens, flow configurations, and dialysates, or alternative methods of sterilization, rinsing, and priming of the dialysers.